KitGuru KitGuru.net – Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards
The last two years have felt like an eternity in the graphics card market, with almost the entire GPU generation facing supply shortages, scalper prices and insatiable demand from cryptocurrency miners. Today, however, heralds the start of something new. The next GPU generation is about to kick off, with Nvidia unveiling its Ada Lovelace architecture and RTX 40-series at GTC last month.
We can now present our review of the flagship RTX 4090, packing 76.3 billion transistors, 16384 CUDA Cores and a boost clock that exceeds 2.5GHz…
| RTX 4090 | RTX 3090 Ti | RTX 3090 | RTX 3080 Ti | RTX 3080 | |
| Process | TSMC 4N | Samsung 8N | Samsung 8N | Samsung 8N | Samsung 8N |
| SMs | 128 | 84 | 82 | 80 | 68 |
| CUDA Cores | 16384 | 10752 | 10496 | 10240 | 8704 |
| Tensor Cores | 512 | 336 | 328 | 320 | 272 |
| RT Cores | 128 | 84 | 82 | 80 | 68 |
| Texture Units | 512 | 336 | 328 | 320 | 272 |
| ROPs | 176 | 112 | 112 | 112 | 96 |
| GPU Boost Clock | 2520 MHz | 1860 MHz | 1695 MHz | 1665 MHz | 1710 MHz |
| Memory Data Rate | 21 Gbps | 21 Gbps | 19.5 Gbps | 19 Gbps | 19 Gbps |
| L2 Cache | 73729 KB | 6144 KB | 6144 KB | 6144 KB | 5120 KB |
| Total Video Memory | 24GB GDDR6X | 24GB GDDR6X | 24GB GDDR6X | 12GB GDDR6X | 10GB GDDR6X |
| Memory Interface | 384-bit | 384-bit | 384-bit | 384-bit | 320-bit |
| Memory Bandwidth | 1008 GB/Sec | 1008 GB/Sec | 936 GB/Sec | 912 GB/Sec | 760 GB/Sec |
| TGP | 450W | 450W | 350W | 350W | 320W |
Unlike the launch of the RTX 30-series back in September 2020, where Nvidia kicked off the new generation with the RTX 3080 and then followed it up with the RTX 3090, this time around it is the big boy being released first – the RTX 4090. Two RTX 4080 SKUs are on the way but won't be landing until next month, and after that… well, your guess is as good as mine.
Still, with the RTX 4090, Nvidia hopes to inject new life into the GPU market. Switching from Samsung 8nm to TSMC 4nm has allowed the manufacturer to vastly increase transistor count, up from 28.3 billion with GA102, to a staggering 76.3 billion with the largest Ada GPU, AD102. Accordingly, core count has gone through the roof, with the RTX 4090 home to 16384 CUDA cores, with 512 Tensor cores and 128 RT cores, with a sizeable increase to the ROP count too, up to 176.
Such a leap forward in process node has also allowed Nvidia to crank the clock speed significantly, with the RTX 4090 now sporting a 2520MHz rated boost clock. Most of the Ampere generation ran at around 1800MHz, so that is really quite a generational improvement.
Memory configuration, however, is almost identical to the RTX 3090 Ti, with 24GB GDDR6X running at 21Gbps, operating over 384-bit interface, totalling memory bandwidth of just over 1TB/s. That said, there has been a substantial upgrade to the L2 cache with the Ada architecture, with the RTX 4090 now offering 73MB, compared to 6MB for GA102.
Power draw also remains at 450W, as per the RTX 3090 Ti, but we are using our updated GPU power testing methodology in this review, so read on for our most detailed power and efficiency testing yet.
To view all pages of this review, click HERE.
